Parts a-g of Problem 1 ask you to assume that whether a premature baby is screened for PDA by ultrasound is NOT independently related to in-hospital mortality. This makes screening a good instrumental variable which is associated with PDA treatment and not otherwise associated with mortality.
Parts h and i, ask you to drop that assumption. The assumption behind parts h and i is that certain characteristics of the baby make both screening and death more likely. In other words, confounding affects the association between screening and death. In parts h and i, the intervention of interest is screening itself, not PDA treatment, and the problem of confounding is addressed by comparing screened and unscreened infants matched on propensity score.
I tried to clarify this in the newly posted version of the problem set.